AVh9)dZddlmZddlmZddlmZej dej dej dej dej d ej d ej d ej d ej d ej dej dej dej ddZejdejdejddejfdZ ejdejdejddejfdZ ejdejdejddejfdZ ejd ejd!ejd"dejfd#Z ejd$ejd%ejd&dejfd'Zejd(ejd)ejd*dejfd+Zy,)-z7Gradients for operators defined in tensor_array_ops.py.)ops) array_ops)tensor_array_ops TensorArrayTensorArrayGradTensorArraySizeTensorArrayClose TensorArrayV2TensorArrayGradV2TensorArraySizeV2TensorArrayCloseV2 TensorArrayV3TensorArrayGradV3TensorArrayGradWithShapeTensorArraySizeV3TensorArrayCloseV3c |jjd}t|Dcgc]\}}|jds|}}}|st d|jddj |d|ddzScc}}w)aIdentify which call to tf.gradients created this gradient op or tensor. TensorArray gradient calls use an accumulator TensorArray object. If multiple gradients are calculated and run in the same session, the multiple gradient nodes may accidentally flow through the same accumulator TensorArray. This double counting breaks the TensorArray gradient flow. The solution is to identify which gradient call this particular TensorArray*Grad is being called in, by looking at the input gradient tensor's name, and create or lookup an accumulator gradient TensorArray associated with this specific call. This solves any confusion and ensures different gradients from the same forward graph get their own accumulators. This function creates the unique label associated with the tf.gradients call that is used to create the gradient TensorArray. Args: op_or_tensor: `Tensor` or `Operation` which is an input to a TensorArray*Grad call. Returns: A python string, the unique label associated with this particular gradients calculation. Raises: ValueError: If not called within a gradients calculation. / gradientszHExpected op/tensor name to start with gradients (excluding scope), got: z. This means that a tf.gradients op with this op in its dependency path has a custom name that does not start with 'gradients'. Please make sure all calls to tf.gradients that have non-empty `name` arguments use names that start with 'gradients'.N)namesplit enumerate startswith ValueErrorjoin) op_or_tensor name_tokensixgrad_poss W/home/dcms/DCMS/lib/python3.12/site-packages/tensorflow/python/ops/tensor_array_grad.py_GetGradSourcer$'s8!!'',+%k2 PDAqall;6Oa P( P   ##$%   +/x|a/0 11Qs BBTensorArrayReadTensorArrayReadV2TensorArrayReadV3opc*|jd}|jd}|jd}|jd}t|}tj|||dj ||}|j ||}dd|jgS) aGradient for TensorArrayRead. Args: op: Forward TensorArrayRead op. grad: Gradient `Tensor` to TensorArrayRead. Returns: A flow `Tensor`, which can be used in control dependencies to force the write of `grad` to the gradient `TensorArray`. rrdtypeFr+handleflowcolocate_with_first_write_callsourcer.N)inputsget_attrr$rrgradwriter.) r(r4r-indexr.r+ grad_sourcegw_gs r#_TensorArrayReadGradr:Ps& 99Q<& ))A,% 1$ ++g %t$+##%TCHJ Kd+ t# chh TensorArrayWriteTensorArrayWriteV2TensorArrayWriteV3c|jd}|jd}|jd}t|}tj|j dd}t j|g5tj|d}dddtj|||dj|| }|j|}dd||gS#1swYJxYw) zGradient for TensorArrayWrite. Args: op: Forward TensorArrayWrite op. flow: Gradient `Tensor` flow to TensorArrayWrite. Returns: A grad `Tensor`, the gradient created in an upstream ReadGrad or PackGrad. rrTflow_out write_barrierNFr,r0) r2r3r$ridentityoutputsrcontrol_dependenciesrrr4read) r(r.r-r6r+r7rAr8r4s r#_TensorArrayWriteGradrGos 99Q<& ))A,% ++c %t$+    1 z :(  +5   dO 4D5##%TCHJ Kd+ $ dD !! 55 4CCTensorArrayGatherTensorArrayGatherV2TensorArrayGatherV3c*|jd}|jd}|jd}|jd}t|}tj|||dj ||}|j ||}dd|jgS) aGradient for TensorArrayGather. Args: op: Forward TensorArrayGather op. grad: Gradient `Tensor` to TensorArrayGather. Returns: A flow `Tensor`, which can be used in control dependencies to force the write of `grad` to the gradient `TensorArray`. rrr*r+Fr,r0N)r2r3r$rrr4scatterr.) r(r4r-indicesr.r+r7r8u_gs r#_TensorArrayGatherGradrPs& 99Q<& IIaL' 1$ ++g %t$+##%TCHJ Kd+ '4 # chh r;TensorArrayScatterTensorArrayScatterV2TensorArrayScatterV3c|jd}|jd}|jd}t|}tj|j dd}t j|g5tj|d}dddtj|||dj|| }|j|}dd||gS#1swYJxYw) zGradient for TensorArrayScatter. Args: op: Forward TensorArrayScatter op. flow: Gradient `Tensor` flow to TensorArrayScatter. Returns: A grad `Tensor`, the gradient created in upstream ReadGrads or PackGrad. rrr@rArBNFr,r0) r2r3r$rrCrDrrErrr4gather) r(r.r-rNr+r7rAr8r4s r#_TensorArrayScatterGradrVs 99Q<& IIaL' ++c %t$+    1 z :(  +5   dO 4D5##%TCHJ Kd+ ' $ dD !! 55rHTensorArrayConcatTensorArrayConcatV2TensorArrayConcatV3c*|jd}|jd}|jd}|jd}t|}t j |||dj ||}|j||} d| jgS) aGradient for TensorArrayConcat. Args: op: Forward TensorArrayConcat op. grad: Gradient `Tensor` to TensorArrayConcat. Returns: A flow `Tensor`, which can be used in control dependencies to force the write of `grad` to the gradient `TensorArray`. rrr+Fr,r0)lengthsN) r2rDr3r$rrr4rr.) r(r4unused_lengths_gradr-r.r[r+r7r8rOs r#_TensorArrayConcatGradr]s& 99Q<& 1$ JJqM' ++g %t$+##%TCHJ Kd+ g&#  r;TensorArraySplitTensorArraySplitV2TensorArraySplitV3c|jd}|jd}t|}tj|j dd}t j|g5tj|d}dddtj|||dj||}|j}d|d|gS#1swYIxYw) zGradient for TensorArraySplit. Args: op: Forward TensorArraySplit op. flow: Gradient `Tensor` flow to TensorArraySplit. Returns: A grad `Tensor`, the gradient created in upstream ReadGrads or PackGrad. rr@rArBNFr,r0) r2r3r$rrCrDrrErrr4concat)r(r.r-r+r7rAr8r4s r#_TensorArraySplitGradrcs 99Q<& ++c %t$+    1 z :(  +5   dO 4D5##%TCHJ Kd+ $ dD !!55s %CC N)__doc__tensorflow.python.frameworkrtensorflow.python.opsrrNotDifferentiabler$RegisterGradient Operationr:rGrPrVr]rcr;r#rks >++2m$'('(()o&)*)**+o&)*01)**+&2R'()*)* S]] ++) 8()*+*+"cmm",,*":)*+,+, s}} --+ 8*+,-,-" "..,"6)*+,+,s}}--+:()*+*+"cmm",,*"r;